Learning region-guided scale-aware feature selection for object detection
- PDF / 3,096,236 Bytes
- 15 Pages / 595.276 x 790.866 pts Page_size
- 1 Downloads / 220 Views
(0123456789().,-volV)(0123456789().,-volV)
ORIGINAL ARTICLE
Learning region-guided scale-aware feature selection for object detection Liu Liu1,2 • Rujing Wang2 • Chengjun Xie2 • Rui Li2 • Fangyuan Wang1,2 • Man Zhou1,2 Yue Teng1,2
•
Received: 8 May 2020 / Accepted: 24 September 2020 Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Scale variation is one of the major challenges in object detection task. Modern region-based object detection architectures often adopt Feature Pyramid Network (FPN) as feature extraction neck to achieve multi-scale feature representation in solving scale variation problem. However, due to the rough feature selection strategy in Region of Interest (RoI) feature extraction step, these methods might not perform well on object detection under strong scale variation. In this work, we are motivated by the limitations of current FPN-based two-stage object detectors and then present a novel module, namely scale-aware feature selective (SAFS) module, that flexibly and adaptively selects feature levels in two-stage object detectors. Specifically, we firstly build the RoI Pyramid in standard FPN structure to extract RoI features from various scale levels. Next, in order to achieve scale-aware mechanism for solving scale variation issue, we develop a novel weighting gate function containing one set of trainable parameters to automatically learn the fusion weight for each RoI feature level, which relieves the limitation of hard feature selection strategy guided by online instance size. Outputs from the RoI features with the learned weights are fused for classification and bounding box regression. Furthermore, we design a multilevel SAFS architecture to obtain different types of RoI feature combinations that ensures our method is more robust to various instance scales. Experimental results show that our SAFS module is very compatible with most of two-stage object detectors and could achieve state-of-the-art results with Average Precision of 48.3 on COCO test-dev and other popular object detection benchmarks. Our code will be made publicly available. Keywords Scale variation Object detection RoI Pyramid Scale-aware feature selective
1 Introduction In modern computer vision field, Convolutional Neural Network (CNN) has shown the high efficiency on automatically extracting powerful features on various visual tasks guided by supervised learning [1–4]. The past decade has witnessed the superior performance when CNN is employed in object detection architectures [5–7]. Among these methods, two-stage approaches are fast becoming a key instrument in generic object detection task due to its & Rujing Wang [email protected] 1
University of Science and Technology of China, Hefei 230026, China
2
Institute of Intelligent Machines, Chinese Academy of Sciences, Hefei 230031, China
high-quality candidate boxes outputted from Region Proposal Network (RPN). Furthermore, Feature Pyramid Network (FPN) [8], introduced as model neck component, can play a vital role in addressing the issue o
Data Loading...
